Eviralnews Repots;
The latest version of artificial intelligence introduced by the OpenAI company has the ability to “think” and its similarity with humans has increased to such an extent that it competes with the best experts in the doctoral degree of various fields.
Eviralnews,; OpenAI's latest AI model, o1, in GPT chat now thinks before responding. This means that the new model is much more advanced than the previous two models and is even more specialized than people who have studied at the doctorate level of a field (PhD).
In such a situation, OpenAI seems to be closing in on us, namely GPT-4o and its still-weak advanced audio mode was unveiled in May, but its improvements are really minor. is The Sora text-to-video model really blew people's minds back in February, but it's not yet available to the public, while several Chinese competitors have offered similar versions.
There is a lot of speculation about what features GPT-5 will have, when it will be launched, and what it will define as AI, but OpenAI has already released a groundbreaking model of GPT with a new division.
Introducing the “O1” model: thinking artificial intelligence
The new GPT model is called “O1” and currently, 100% of GPT chat user accounts are activated as an optional option. The GPT-4O is a versatile and versatile model that covers the most tasks, but the new O1 model does specialized work.
The special feature of the “U1” model is its reasoning power, that is, it “thinks” a little before answering.
It may be tempting to humanize language models. It is true that these machines are not human, but since they are trained on human data, their similarities to humans often scare us! In this case, the “O1” model has a much higher performance than the previous models in doing difficult tasks, this model divides a large task into smaller parts and carefully evaluates the task and its assumptions before giving you an answer. checks
GPT-4O usually responds instantly, writes code, or generates images or text. “Model O1” might make you pause for a while to think. Of course, not much time, maybe 10 or 20 seconds, because it is supposed to solve much more complex problems.
In fact, the more this new model can think, the better, but today's models only think for a few seconds. OpenEye says that future versions will likely think for hours or even days to come up with a logical answer, meaning they'll explore huge and complex problems, experiment with different solutions, and finally come up with an answer.
Limitations of “Model O1”
The Model O1 is now available in Preview and Mini models. They can write code and run it, but these features are beta previews and lack a few important components:
You cannot upload files to them
• They don't have access to your GPT-4O memory or custom requests, so they don't know anything about you.
• They do not have access to web information after the end of their training course (October 2023).
In general, “GPT-4O” works much more useful in any case that needs to upload a file or access the web.
Advantages of “Model O1”
“Model O1” shows remarkable effectiveness, especially in scientific subjects. This model achieved a perfect score in the Open Eye coding test, which shows its strong coding capabilities. The model also performed exceptionally well on PhD-level questions, especially outperforming PhD-level physicists in physics. But he could not perform better than biology and chemistry experts.
This model has also shown significant improvements in mathematics, especially in tough competitions such as the 2024 American Call for Mathematics Competitions (AIME 2024).
Despite all the strengths of this model, OpenAI continues to hire human engineers, because the nuances of teamwork, creativity, and problems that humans can solve are still not up to AI, and there is room for serious improvement in some areas. has
This report has been translated into Farsi from the New Atlas website.
